# Keras

## Sequential详解

```python
# 导入
from keras.models import Sequential
from keras.layers import Dense, Activation
# Sequentia的两种使用方法
# 1 直接堆栈
model = Sequential([
    Dense(32, input_shape=(784,)),
    Activation('relu'),
    Dense(10),
    Activation('softmax'),
])

# 2 .add()方法堆栈
model = Sequential()
model.add(Dense(32, input_dim=784))
model.add(Activation('relu'))
model.add(Dense(64, activation='relu'))
```

## Dense详解

```python
keras.layers.Dense(units, 
				  activation=None, 
				  use_bias=True, 
				  kernel_initializer='glorot_uniform', 
				  bias_initializer='zeros', 
				  kernel_regularizer=None, 
				  bias_regularizer=None, 
			      activity_regularizer=None, 
				  kernel_constraint=None, 
				  bias_constraint=None)
# 定义第一层时需要指定数据输入的形状，有参数input_dim
# input_dim也可以用input_shape来代替，如mnist的第一层输入input_dim=784,或者input_shape=(784,)
```

### 参数说明：

- units: 该层有几个神经元

- activation: 该层使用的激活函数

- use_bias: 是否添加偏置项

- kernel_initializer: 权重初始化方法

- bias_initializer: 偏置值初始化方法

- kernel_regularizer: 权重规范化函数

- bias_regularizer: 偏置值规范化方法

- activity_regularizer: 输出的规范化方法

- kernel_constraint: 权重变化限制函数

- bias_constraint: 偏置值变化限制函数

### 用法

```python
from keras.layers import Dense, Activation

Dense(512, activation='sigmoid', input_dim=2, use_bias=True)
# 这里定义了一个有512个节点，使用sigmoid激活函数的神经层
```


<!-- markdown for nginx, see https://phus.lu -->
<script>
!function(){
	var dom = {
		element: null,
		get: function (o) {
			var obj = Object.create(this)
			obj.element = (typeof o == "object") ? o : document.createElement(o)
			return obj
		},
		add: function (o) {
			var obj = dom.get(o)
			this.element.appendChild(obj.element)
			return obj
		},
		text: function (t) {
			this.element.appendChild(document.createTextNode(t))
			return this
		},
		attr: function (k, v) {
			this.element.setAttribute(k, v)
			return this
		}
	}

	if (!document.head) return
	head = dom.get(document.head)
	head.add('meta').attr('charset', 'utf-8')
	head.add('meta').attr('name', 'viewport').attr('content', 'width=device-width,initial-scale=1')

	if (!document.body) {
		document.write(["<div class=\"container\">",
		"<h3>nginx.conf</h3>",
		"<textarea rows=6 cols=50>",
		"# download markdown.html to /wwwroot",
		"location ~ \\.md$ {",
		"    default_type text/html;",
		"    add_after_body /markdown.html;",
		"}",
		"</textarea>",
		"</div>"].join("\n"))
		return
	}

	var bodytext = document.body.innerHTML
	document.body.innerHTML = ''

	div = dom.get('div').attr('class', 'container')

	div.add('script').attr('src', 'https://cdnjs.cloudflare.com/ajax/libs/marked/0.5.2/marked.min.js').attr('integrity', 'sha256-zFUosuESzULu5P+SZdjRRtBZR8+1u5RZDlbt3Q5KL8U=').attr('crossorigin', 'anonymous')
	div.add('link').attr('rel', 'stylesheet').attr('href', 'https://cdnjs.cloudflare.com/ajax/libs/github-markdown-css/5.1.0/github-markdown-light.min.css').attr('integrity', 'sha256-WQx0Y6LLZeGv3V3NVVge+YIL5MIOt45RPuFdkyNodls=').attr('crossorigin', 'anonymous')

	title = decodeURIComponent(document.location.pathname.replace(/.*\//, '').replace(/\.html$/, ''))
	document.title = '《' + title.replace(/\.md$/, '') + '》'
	tbody = div.add('table').add('tbody')
	tbody.add('tr').add('th').text(title).attr('class', 'octicon-book')
	tbody.add('tr').add('td').add('div').attr('id', 'readme').attr('class', 'markdown-body')

	wait = function (name, callback) {
		var interval = 10; // ms
		window.setTimeout(function() {
			if (window[name]) {
				callback(window[name])
			} else {
				window.setTimeout(arguments.callee, interval)
			}
		}, interval)
	}
	wait('marked', function() {
		document.getElementById("readme").innerHTML = marked.parse(bodytext)
	})

	document.body.appendChild(div.element)
}()
</script>

<style>
body {
	margin: 0;
	font-family: "ubuntu", "Tahoma", "Microsoft YaHei", Arial, Serif;
}
.markdown-body {
	float: left;
	font-family: "ubuntu", "Tahoma", "Microsoft YaHei", Arial, Serif;
}
.container {
	padding-right: 15px;
	padding-left: 15px;
	margin-right: auto;
	margin-left: auto;
}
@media (min-width: 768px) {
	.container {
		max-width: 80%;
	}
}
@media (min-width: 992px) {
	.container {
		max-width: 70%;
	}
}
@media (min-width: 1200px) {
	.container {
		max-width: 60%;
	}
}
table {
	width: 100%;
	max-width: 100%;
	margin-bottom: 20px;
	border: 1px solid #ddd;
	padding: 0;
	border-collapse: collapse;
}
table th {
	font-size: 14px;
}
table tr {
	border: 1px solid #ddd;
	padding: 5px;
}
table th, table td {
	border: 1px solid #ddd;
	font-size: 14px;
	line-height: 20px;
	padding: 3px;
	text-align: left;
}
.octicon-book {
	background-position: center left;
	background-repeat: no-repeat;
	padding-left: 20px;
	background-image: url("data:image/svg+xml;charset=utf8,%3Csvg xmlns='http://www.w3.org/2000/svg' width='16' height='16' viewBox='0 0 16 16'%3E%3Cpath d='M3,5 L7,5 L7,6 L3,6 L3,5 L3,5 Z M3,8 L7,8 L7,7 L3,7 L3,8 L3,8 Z M3,10 L7,10 L7,9 L3,9 L3,10 L3,10 Z M14,5 L10,5 L10,6 L14,6 L14,5 L14,5 Z M14,7 L10,7 L10,8 L14,8 L14,7 L14,7 Z M14,9 L10,9 L10,10 L14,10 L14,9 L14,9 Z M16,3 L16,12 C16,12.55 15.55,13 15,13 L9.5,13 L8.5,14 L7.5,13 L2,13 C1.45,13 1,12.55 1,12 L1,3 C1,2.45 1.45,2 2,2 L7.5,2 L8.5,3 L9.5,2 L15,2 C15.55,2 16,2.45 16,3 L16,3 Z M8,3.5 L7.5,3 L2,3 L2,12 L8,12 L8,3.5 L8,3.5 Z M15,3 L9.5,3 L9,3.5 L9,12 L15,12 L15,3 L15,3 Z' /%3E%3C/svg%3E");
}
</style>